♻️ refactor new datacenter workflow (#3964)

Author: thomas-lebeau

.gitlab/deploy-auto.yml

@@ -17,7 +17,7 @@ stages:
     - VERSION=$(node -p -e "require('./lerna.json').version")
     - yarn
     - yarn build:bundle
-    - node ./scripts/deploy/deploy-prod-dc.ts v${VERSION%%.*} $UPLOAD_PATH true
+    - node ./scripts/deploy/deploy-prod-dc.ts v${VERSION%%.*} $UPLOAD_PATH --check-monitors
 
 step-1_deploy-prod-minor-dcs:
   when: manual
@@ -26,7 +26,7 @@ step-1_deploy-prod-minor-dcs:
     - .base-configuration
     - .deploy-prod
   variables:
-    UPLOAD_PATH: us3,us5,ap1,ap2
+    UPLOAD_PATH: minor-dcs
 
 step-2_deploy-prod-eu1:
   needs:

.gitlab/deploy-manual.yml

@@ -19,14 +19,14 @@ stages:
     - VERSION=$(node -p -e "require('./lerna.json').version")
     - yarn
     - yarn build:bundle
-    - node ./scripts/deploy/deploy-prod-dc.ts v${VERSION%%.*} $UPLOAD_PATH false
+    - node ./scripts/deploy/deploy-prod-dc.ts v${VERSION%%.*} $UPLOAD_PATH --no-check-monitors
 
 step-1_deploy-prod-minor-dcs:
   extends:
     - .base-configuration
     - .deploy-prod
   variables:
-    UPLOAD_PATH: us3,us5,ap1,ap2
+    UPLOAD_PATH: minor-dcs
 
 step-2_deploy-prod-eu1:
   extends:
@@ -79,6 +79,13 @@ step-7_create-github-release:
     - yarn
     - node scripts/release/create-github-release.ts
 
+# This step is used to deploy the SDK to a new datacenter.
+# the `UPLOAD_PATH` variable needs to be provided as an argument when starting the manual job
+optional_step-deploy-to-new-datacenter:
+  extends:
+    - .base-configuration
+    - .deploy-prod
+
 ########################################################################################################################
 # Notify
 ########################################################################################################################

scripts/deploy/check-monitors.ts

@@ -5,30 +5,15 @@
  */
 import { printLog, runMain, fetchHandlingError } from '../lib/executionUtils.ts'
 import { getTelemetryOrgApiKey, getTelemetryOrgApplicationKey } from '../lib/secrets.ts'
-import { siteByDatacenter } from '../lib/datadogSites.ts'
-
-interface MonitorsByDatacenter {
-  [datacenter: string]: number[]
-}
-
-const monitorIdsByDatacenter: MonitorsByDatacenter = {
-  us1: [72055549, 68975047, 110519972],
-  eu1: [5855803, 5663834, 9896387],
-  us3: [164368, 160677, 329066],
-  us5: [22388, 20646, 96049],
-  ap1: [858, 859, 2757030],
-  ap2: [1234, 1235, 1236],
-}
-
-type Datacenter = keyof typeof siteByDatacenter
+import { monitorIdsByDatacenter, siteByDatacenter } from '../lib/datacenter.ts'
 
 interface MonitorStatus {
   id: number
   name: string
   overall_state: string
 }
 
-const datacenters = process.argv[2].split(',') as Datacenter[]
+const datacenters = process.argv[2].split(',')
 
 runMain(async () => {
   for (const datacenter of datacenters) {

scripts/deploy/deploy-prod-dc.ts

@@ -1,5 +1,7 @@
+import { parseArgs } from 'node:util'
 import { printLog, runMain, timeout } from '../lib/executionUtils.ts'
 import { command } from '../lib/command.ts'
+import { siteByDatacenter } from '../lib/datacenter.ts'
 
 /**
  * Orchestrate the deployments of the artifacts for specific DCs
@@ -10,19 +12,35 @@ const ONE_MINUTE_IN_SECOND = 60
 const GATE_DURATION = 30 * ONE_MINUTE_IN_SECOND
 const GATE_INTERVAL = ONE_MINUTE_IN_SECOND
 
-const version: string = process.argv[2]
-const uploadPath: string = process.argv[3]
-const withMonitorChecks: boolean = process.argv[4] === 'true'
+const {
+  values: { 'check-monitors': checkMonitors },
+  positionals,
+} = parseArgs({
+  allowPositionals: true,
+  allowNegative: true,
+  options: {
+    'check-monitors': {
+      type: 'boolean',
+    },
+  },
+})
+
+const version = positionals[0]
+const uploadPath = positionals[1] === 'minor-dcs' ? getAllMinorDcs().join(',') : positionals[1]
+
+if (!uploadPath) {
+  throw new Error('UPLOAD_PATH argument is required')
+}
 
 runMain(async () => {
-  if (withMonitorChecks) {
+  if (checkMonitors) {
     command`node ./scripts/deploy/check-monitors.ts ${uploadPath}`.withLogs().run()
   }
 
   command`node ./scripts/deploy/deploy.ts prod ${version} ${uploadPath}`.withLogs().run()
   command`node ./scripts/deploy/upload-source-maps.ts ${version} ${uploadPath}`.withLogs().run()
 
-  if (withMonitorChecks && uploadPath !== 'root') {
+  if (checkMonitors && uploadPath !== 'root') {
     await gateMonitors(uploadPath)
   }
 })
@@ -36,3 +54,12 @@ async function gateMonitors(uploadPath: string): Promise<void> {
   }
   printLog() // new line
 }
+
+// Major DCs are the ones that are deployed last.
+// They have their own step jobs in `deploy-manual.yml` and `deploy-auto.yml`.
+const MAJOR_DCS = ['root', 'us1', 'eu1']
+
+// Minor DCs are all the DCs from `siteByDatacenter` that are not in `MAJOR_DCS`.
+function getAllMinorDcs(): string[] {
+  return Object.keys(siteByDatacenter).filter((dc) => !MAJOR_DCS.includes(dc))
+}

scripts/deploy/upload-source-maps.spec.ts

@@ -1,7 +1,7 @@
 import assert from 'node:assert/strict'
 import path from 'node:path'
 import { beforeEach, before, describe, it, mock } from 'node:test'
-import { siteByDatacenter } from '../lib/datadogSites.ts'
+import { siteByDatacenter } from '../lib/datacenter.ts'
 import { mockModule, mockCommandImplementation, replaceChunkHashes } from './lib/testHelpers.ts'
 
 const FAKE_API_KEY = 'FAKE_API_KEY'

scripts/deploy/upload-source-maps.ts

@@ -3,7 +3,7 @@ import { printLog, runMain } from '../lib/executionUtils.ts'
 import { command } from '../lib/command.ts'
 import { getBuildEnvValue } from '../lib/buildEnv.ts'
 import { getTelemetryOrgApiKey } from '../lib/secrets.ts'
-import { siteByDatacenter } from '../lib/datadogSites.ts'
+import { siteByDatacenter } from '../lib/datacenter.ts'
 import { forEachFile } from '../lib/filesUtils.ts'
 import { buildRootUploadPath, buildDatacenterUploadPath, buildBundleFolder, packages } from './lib/deploymentUtils.ts'
 

scripts/lib/datacenter.ts

@@ -0,0 +1,24 @@
+export const siteByDatacenter: Record<string, string> = {
+  us1: 'datadoghq.com',
+  eu1: 'datadoghq.eu',
+  us3: 'us3.datadoghq.com',
+  us5: 'us5.datadoghq.com',
+  ap1: 'ap1.datadoghq.com',
+  ap2: 'ap2.datadoghq.com',
+  pr00test: 'pr00test.staging.dog',
+}
+
+/**
+ * Each datacenter has 3 monitor IDs:
+ * - Telemetry errors
+ * - Telemetry errors on specific org
+ * - Telemetry errors on specific message
+ */
+export const monitorIdsByDatacenter: Record<string, [number, number, number]> = {
+  us1: [72055549, 68975047, 110519972],
+  eu1: [5855803, 5663834, 9896387],
+  us3: [164368, 160677, 329066],
+  us5: [22388, 20646, 96049],
+  ap1: [858, 859, 2757030],
+  ap2: [1234, 1235, 1236],
+}